Systems and methods to reduce data and complexity in neural signal processing chain

Abstract

Described herein are systems and methods for reducing the size of data payloads delivered to downstream processing from a raw series of biological sensor recordings. In one variation, the system comprises a low-power hardware architecture that combines serially sampled neural signal data with a transformation matrix (TFM) using a novel systolic random-logic-macro (RLM) array.

The invention claimed is: 
     
       1. A method for detecting neural activity, the method comprising:
 acquiring neural signal data from a plurality of M sensors to generate a data channel vector having M data channels; 
 generating a transformation matrix (TFM) comprising an M×N matrix of matrix values, wherein the matrix values are determined by:
 identifying high-noise data channels in the data channel vector with high-noise neural signal data and assigning a zero value to the matrix values in rows of the TFM that correspond to the identified high-noise data channels; 
 identifying correlated data channels in the data channel vector that each have correlated neural signal data that is correlated one with neural signal data in one or more other data channels in the data channel vector; and 
 iterating through the matrix values of the TFM to compute non-zero matrix values for rows of the TFM that correspond to the identified correlated data channels, wherein the non-zero matrix values of the TFM combine the neural signal data of the correlated data channels; and 
 
 configuring an array of individually addressable logic blocks to have M×N active logic blocks, where each of the addressable logic blocks comprises a memory element that stores a matrix value of the TFM corresponding to each of the addressable logic blocks; and 
 generating an output data channel vector having N data elements by multiplying the data channel vector with the TFM using the M×N array of logic blocks. 
 
     
     
       2. The method of  claim 1 , further comprising displaying the matrix values of the TFM on a monitor. 
     
     
       3. The method of  claim 1 , wherein the logic block array is located in an implantable device and wherein the method further comprises transmitting the output data channel vector from the implantable device to an external controller. 
     
     
       4. The method of  claim 3 , wherein the plurality of M sensors are configured to be attached to the implantable device and are implantable into or in contact with brain tissue, and the data channel vector is configured to be transmitted from the implantable device to a processor of the external controller, and wherein the TFM is generated using the processor of the external controller and transmitted to the implantable device. 
     
     
       5. The method of  claim 1 , wherein configuring the logic block array comprises providing a gating enable signal to the M×N active logic blocks. 
     
     
       6. The method of  claim 1 , wherein configuring the logic block array comprises providing a gating enable signal to a number of active logic blocks that correspond with the number of non-zero matrix values of the TFM. 
     
     
       7. The method of  claim 1 , wherein configuring the logic block array comprises providing a gating power signal to the M×N active logic blocks. 
     
     
       8. The method of  claim 1 , wherein the identifying of the correlated data channels with the correlated neural signal data comprises calculating a correlation coefficient between pairs of data channels in the data channel vector and clustering the pairs of data channels according to the calculated correlation coefficients. 
     
     
       9. The method of  claim 1 , wherein the identifying of the high-noise data channels with the high-noise neural signal data comprises comparing a magnitude of predetermined spectral components of the neural signal data of the M data channels and identifying data channels of the M data channels that have a magnitude of predetermined spectral components greater than a predetermined threshold value, and the generating of the TFM comprises assigning the zero value to the matrix values that correspond with the the high-noise data channels. 
     
     
       10. The method of  claim 1 , wherein generating the TFM further comprises identifying valid data channels with low-noise neural signal data that have noise levels below a noise threshold, and assigning the non-zero value to the matrix values that correspond with the valid data channels. 
     
     
       11. The method of  claim 1 , wherein generating the TFM further comprises iterating through the matrix values of the TFM such that multiplying the data channel vector with the TFM selects data channels of the M data channels that have neural signal data pertaining to a neural activity characteristic. 
     
     
       12. The method of  claim 11 , wherein the neural activity characteristic comprises a frequency component of the neural signal data. 
     
     
       13. The method of  claim 12 , wherein the frequency component comprises neural signal data in the beta frequency band (15-30 Hz). 
     
     
       14. The method of  claim 1 , further comprising generating a data validity flag for each data channel of the data channel vector indicating whether each data channel contains neural signal data having noise levels that exceed a noise threshold. 
     
     
       15. The method of  claim 1 , wherein each logic block of the logic block array comprises a multiplication circuit and an addition circuit. 
     
     
       16. The method of  claim 15 , wherein the logic block array is configured as a systolic matrix multiplier. 
     
     
       17. The method of  claim 1 , wherein the M sensors comprise M electrodes connected to an analog-to-digital converter (ADC), and the method further comprises generating the TFM if an impedance of an electrode exceeds a pre-determined threshold or if neural signal data measured by the electrodes is outside of an operating range of the ADC. 
     
     
       18. The method of  claim 17 , wherein the method comprises generating the TFM when the neural signal data is clipped by the ADC. 
     
     
       19. The method of  claim 1 , wherein the TFM is bipolar-type matrix, a Laplacian-type matrix, a global weighted average matrix, or a local weighted average matrix. 
     
     
       20. The method of  claim 1 , wherein generating the TFM occurs during a calibration session after the sensors have been implanted in or on brain tissue. 
     
     
       21. The method of  claim 1 , wherein the sensors are implanted into brain tissue, and generating the TFM occurs when one or more of the plurality of implanted sensors has shifted position within the brain tissue. 
     
     
       22. The method of  claim 1 , wherein the sensors are implanted into brain tissue, and generating the TFM occurs when trauma to the brain tissue has been detected. 
     
     
       23. The method of  claim 1 , wherein a first sensor has been implanted into or in contact with a first functional brain region and a second sensor has been implanted into or in contact with a second functional brain region that is different from the first functional brain region. 
     
     
       24. The method of  claim 1 , further comprising transmitting the output data channel vector to an external controller and reconstructing an approximation of the data channel vector using the external controller by multiplying the output data channel vector with an inverse of the TFM. 
     
     
       25. The method of  claim 1 , wherein N<M and the output data channel vector is a reduced data channel vector. 
     
     
       26. A method for acquiring and processing neural signal data acquired by a measurement system, the method comprising:
 acquiring neural signal data from a plurality M of sensors in contact with neural tissue, wherein the neural signal data forms a data channel vector having M channel elements; 
 multiplying the data channel vector with a M×N transformation matrix (TFM) of data channel values to remove neural signal data redundancies and noise artifacts, wherein the product is a reduced data vector having N data elements, wherein N<=M; and 
 transmitting the reduced data vector to an external controller, 
 wherein multiplying the data channel vector with the TFM comprises performing systolic matrix multiplication using a logic block array located inside an implanted measurement system controller, wherein the logic block array comprises one logic block per non-zero data channel value of the TFM and each logic block comprises a memory element that stores a non-zero data channel value of the TFM. 
 
     
     
       27. The method of  claim 26 , further comprising generating a data validity flag for each data channel of the data channel vector indicating whether a data channel contains neural signal data having noise levels that exceed a noise threshold. 
     
     
       28. The method of  claim 27 , wherein a data validity flag is generated if the neural signal data corresponds to a malfunctioning activity sensor and/or has signal noise levels that exceed a predetermined noise threshold. 
     
     
       29. The method of  claim 26 , wherein performing systolic matrix multiplication comprises sequentially activating logic blocks to sequentially multiply data channel values of the TFM with the channel elements of the data channel vector, and sequentially deactivating logic blocks in synchrony with a system clock signal.